Blurred image restoration method based on unsupervised generative adversarial network

Abstract

The invention relates to a blurred image restoration method based on an unsupervised generative adversarial network, and the method comprises the steps: employing two generators GA and GB to carry out the modeling of a restoration problem of a blurred image and a blurring degradation process of a clear image under the framework of a dual generative adversarial network; introducing a discriminator to cooperate with a corresponding generator to carry out adversarial training, wherein a discriminator DA is used to determine whether an image generated by the GA is clear, and a discriminator DB is used to determine whether an image generated by the GB is fuzzy; meanwhile, introducing L2 pixel reconstruction loss and perception loss to carry out modeling on a target loss function; finally, training the network by using a non-paired fuzzy-clear image data set, updating network parameters by using an Adam optimizer, and finally obtaining a local optimal solution of the model. Compared with a traditional blurred image restoration method, the method has the advantages that the workload of data set construction is greatly reduced, and the method has wide practical value and application prospect.

Description

technical field [0001] The invention relates to the field of computer digital image processing, in particular to a blurred image restoration method based on an unsupervised generation confrontation network. Background technique [0002] In daily photography, space remote sensing observation, medical image detection and other application fields, the image is often blurred due to the relative motion between the target object and the imaging system, which affects the detail resolution of the image and reduces its use value. Therefore, it is particularly important to design a corresponding blurred image restoration algorithm to improve the resolution of the image. [0003] Traditional blurred image restoration methods usually estimate the blur kernel and deconvolve the blurred image to obtain a clear image. However, the blur kernel estimated by such methods usually cannot represent the real blur and is sensitive to noise. In addition, currently popular supervised learning netwo...

AI Technical Summary

Problems solved by technology

[0003] Traditional blurred image restoration methods generally obtain a clear image by deconvolution with the blurred image after estimating the blurred kernel, but the blurred kernel estimated by this method usually cannot represent the real blurring situation and is sensitive to noise.

In addition, currently popular supervised learning networks based on conditional generative adversarial networks are often limited to training data sets. Collecting a large number of supervised fuzzy-clear training data sets is time-consuming and laborious, and is prone to problems of insufficient network generalization.

Images